Hypothesis Tests 2 – Solutions

STAT-UB.0103 – Statistics for Business Control and Regression Models

Tests on the population mean (known variance)

1. Is a soda-dispensing machine performing according to specification? Pepsi’s dispensing ma-
chine is designed to fill bottles with exactly 2 liters of their product. To test if the machine is
performing according to specification, we collect a sample of 100 “2-liter” bottles. The average
quantity contained in the sample bottles is x̄ = 1.985 liters. The (population) standard devia-
tion of the fill is known to be 0.05. Test whether the machine is in control, at the 5% level of
significance.
(a) What are the null and alternative hypotheses?

Solution: The population we are interested in is all bottles filled by the machine (past
and future); the mean fill level of the population is µ, and the standard deviation of
the fill level of the population is σ = 0.05. Our null and alternative hypotheses are as
follows:

H0 : µ = 2
Ha : µ 6= 2.

The nominal (null) value of the population mean is µ0 = 2.

(b) What is the test statistic?

Solution: Our test statistic is based on the sample of n = 100 bottles. Let x̄ denote
the mean fill level from this sample. Our test statistic is
x̄ − µ0 x̄ − 2 x̄ − 2
z= √ = √ = .
σ/ n (.05)/ 100 (.005)

(c) What is the rejection region?

Solution: Since this is a 2-sided alternative, we reject for large values of |z|, specif-
ically when |z| > zα/2 . Since we are testing at the 5% significance level, we have
α = .05 and zα/2 = 1.96 ≈ 2. We reject H0 when

|z| > 2.

(d) What assumptions are you making?

Solution: We are assuming that the sample bottles were drawn independently with-
out bias from the population.
(e) What is the result of the test?

Solution: The observed value of the test statistic is

1.985 − 2
= −3.
.005
Since this value is in the rejection region, we reject H0 at significance level 5%.

Page 2
2. Before Facebook’s recent redesign, the mean number of ad clicks per day was 100K, and the
standard deviation was 35K. In the 49 days after the redesign, the mean number of ad clicks
per day was 105K. Is there significant evidence that the redesign affected ad clicks? Perform a
test at the 5% level, assuming the redesign did not affect the population standard deviation.
(a) What are the null and alternative hypotheses?

Solution: There are two populations of interest here. The first population is the
daily ad clicks (past and future), in a (hypothetical) world where Facebook does not
redesign their cite. The second population is the daily ad clicks (past and future) in
the world where Facebook redesigns their cite. Let µ0 be the mean clicks per day in
the first population, and µ be the mean clicks per day. Assume that the standard
deviation of the number of clicks per day is the same for both populations, σ.
We use “thousands of clicks” as the units for all relevant quantities. The problem tells
us that µ0 = 100 and σ = 35.
The null hypothesis is that there is no difference between the two populations (i.e., the
redesign had no effect); the alternative hypothesis is that the two populations have
different means:

H0 : µ = 100
Ha : µ 6= 100

(b) What is the test statistic?

Solution: The test statistic is based on the sample, the observed clicks in the n = 49
days after the redesign. Let x̄ denote the mean clicks per day in the sample. Our test
statistic is
x̄ − µ0 x̄ − 100 x̄ − 100
z= √ = √ = .
σ/ n 35/ 49 5

(c) What is the rejection region?

Solution: Since this is a 2-sided test, we reject when |z| is large, specifically when
|z| > zα/2 . The problem asks for a test at level 5%, so α = .05 and zα/2 = 1.96 ≈ 2.
We will reject H0 if |z| > zα/2 , i.e.
|z| > 2.

(d) What assumptions are you making?

Solution: We need to assume that the sample elements were drawn independently
from the population. (The population we refer to is the world in which Facebook does
the website redesign, not the hypothetical world where they don’t.)
Admittedly, the assumption does not hold, since there is strong selection bias in the
sample: we are sampling days right after the website redesign with higher probability

Page 3
 than days far into the future. For example, we have no chance of sampling a day three
years into the future.
Since the assumption does not hold, we are in a bit of an awkward position. The
hypothesis test may not be valid. One way in which it may not be valid is the following:
it usually takes people a few weeks to adjust to a website redesign, so the ad click
behavior in our sample of 49 days may not be representative of all future ad click
behavior.

(e) What is the result of the test?

Solution: Despite the caveats mentioned in part (d), we will proceed with the test.
Using the observed value x̄ = 105, we compute
105 − 100
z= = 1.
5
Since |z| ≤ 2, the statistic is not in the rejection region, so we do not reject H0 at
significance level 5%.

Page 4
Tests on the population mean (unknown variance)

3. This is a variation of problem 1. A certain population has unknown mean µ and unknown
standard deviation σ. You want test (at significance level 5%) the null hypothesis H0 : µ = 2
against the alternative Ha : µ 6= 2. To this end, you collect a sample of size n = 100. The
sample mean is x̄ = 1.985 and the sample standard deviation is s = 0.5.
(a) What is the test statistic?

Solution:
x̄ − µ0 x̄ − 2 x̄ − 2
t= √ = √ = .
s/ n (.05)/ 100 (.005)

(b) What is the rejection region?

Solution: Since this is a 2-sided alternative, we reject for large values of |t|, specifically
when |t| > tα/2 . We have n − 1 = 99 degrees of freedom. Since we are testing at the
5% significance level, we have α = .05 and tα/2 ≈ 2. We reject H0 when

|t| > 2.

(c) What assumptions are you making?

Solution: We are assuming that the samples are drawn independently from the pop-
ulation.

(d) What is the result of the test?

Solution: The observed value of the test statistic is

1.985 − 2
= −3.
.005
Since this value is in the rejection region, we reject H0 at significance level 5%.

(e) How is this problem different from problem ???

Solution: The population standard deviation is unknown, so we use a t statistic

instead of a z statistic.

Page 5
One-sided alternatives

4. The average nicotine content of a brand of cigarettes must be less than 0.5 mg for it to qualify
as a Low Nicotine brand. The manufacturer of Lucky Strikes Cigarettes claims that it is a Low
Nicotine brand. To test this claim, the FDA takes a random sample of 20 cigarettes (one pack)
of Lucky Strikes. They find an average nicotine content of 0.4 mg, with a sample standard
deviation of 0.2 mg. Test the manufacturer’s claim, at the 1% level of significance. Assume
that the nicotine measurements are normally distributed.
(a) What are the population and the sample?

Solution: The population is the nicotine contents of all Low Nicotine cigarettes man-
ufactured by Lucky Strike. The sample is the n = 20 sampled cigarettes.

(b) What are the null and alternative hypotheses?

Solution: Let µ be the average nicotine contents (in mg) of all cigarettes in the
population.

H0 : µ = 0.5
Ha : µ > 0.5

(c) What is the test statistic?

Solution: Since the population standard devation is unknown, we use a t statistic:

X̄ − 0.2
T = √
S/ n

(d) What is the rejection region?

Solution: We reject H0 if T > tα for α = 0.01 and n − 1 = 19 degrees of freedom.

That is, we reject H0 if T > 2.539.

(e) What assumptions are you making?

Solution: We assume that the sample is unbiased. Also, since n < 30, we need to
assume that the population is normally distributed.

(f) What is the result of the test?

Solution: The observed test statistic is

0.4 − 0.5
t= √ = −2.236.
0.2/ 20

Page 6
Since this is not in the rejection region, we do not reject the null hypothesis.

Page 7